Working memory (WM) is a fundamental component of higher cognitive function that enables the persistent encoding of information during movement planning and decision-making. Dysfunction of WM has been linked to schizophrenia, attention deficit disorder, and age-related cognitive decline. In humans and monkeys, the prefrontal cortex (PFC) functions as the hub of a distributed network of brain regions that collectively supports WM, and damage to PFC abolishes WM function entirely. Therefore, understanding the mechanisms of WM function in PFC may be helpful to understanding the etiology of WM-related disorders in humans. Primate WM function has been linked to neurons in PFC whose firing rates remain persistently elevated during a delay period after the termination of a sensory stimulus and preceding a motor response. The mechanisms of persistent activity in PFC remain unknown, but theoretical work presents two competing hypotheses for how network interactions in PFC may give rise to persistent activity during WM. The attractor hypothesis proposes that mutually excitatory interactions produce a temporally invariant mode of network activity that persists after sensory input terminates and is stable in time. By contrast, the feed forward integrator hypothesis proposes that persistent activity results from the transient flow of sensory input through a sequence of modules. WM function then arises from the summed, time-varying outputs from these modules, approximating a step function. Computational models based on both hypotheses reproduce the firing rate responses of isolated PFC neurons during a spatial WM task. However, the experimental data needed to properly distinguish between these two competing frameworks is currently lacking. This proposal combines multiple-electrode recordings, behavioral manipulations and computational techniques to test three opposing predictions made by attractor and feed-forward models about firing rate dynamics and correlated variability during spatial WM. In all experiments, PFC neurons will be isolated and their responses recorded from an electrode array of up to 32 electrodes in monkeys performing a memory-guided oculomotor delayed response (ODR) task. Aim 1 will test conflicting model predictions about correlated variability among pairs of persistently active neurons during WM, by analyzing spike count correlations in paired single unit recordings during the memory ODR task. Aim 2 will test conflicting model predictions about the response of persistently active neurons to sustained visual stimulation, by analyzing single unit firing rate changes over time during a visually-guided ODR task. Finally, Aim 3 will test conflicting model predictions about the response of persistently active neurons to distracter stimuli during WM, by studying differences between single unit firing rates before and after a briefly flashed distracter stimulus. The result of these experiments will improve our understanding of WM function by supporting an existing model or pointing toward a new model of persistent activity generation in PFC.